Weekly signals on how AI is reshaping healthcare, business, leadership, education, and the future of professional work. What changed, why it matters, and the ReasonFirst read.
◆ Signal of the week
LeadershipMIT Sloan Management ReviewJune 9, 2026
CIOs Name Skill Erosion as a Real AI Risk, Not a Theoretical One
At the 2026 MIT Sloan CIO Symposium, technology and business leaders identified the degradation of critical thinking as a concrete organizational risk emerging from AI workflow integration. The concern is not that AI performs tasks poorly, but that humans are losing the capacity to evaluate, question, and override what AI produces.
Why it matters
Healthcare and education leaders who deploy AI to support clinical or academic workflows need to understand that efficiency gains and cognitive atrophy can happen simultaneously and in the same team.
The ReasonFirst take
The real question is not whether AI erodes critical thinking in the abstract, but whether your organization has any feedback loop that would even detect it before the skill is gone.
A state audit of Medicaid records found that pharmacy benefit managers are using complex claims-processing methods that obscure actual drug costs and result in taxpayer overcharges. The findings add to a growing body of evidence that PBM contracting structures lack the transparency needed for meaningful oversight.
Why it matters
Clinician-leaders and health system administrators who negotiate with PBMs or advise on formulary design need to understand that the pricing opacity is not incidental but structural.
The ReasonFirst take
When an independent audit has to reverse-engineer drug costs from Medicaid claims data just to see what was actually paid, that is not a compliance problem waiting to be fixed with better contracts; it is a system designed to resist accountability, and the policy response needs to match that reality.
Former employees of LifeMD, a telehealth company promoted in connection with Novo Nordisk's obesity drug programs, allege that clinicians were pressured to prescribe faster with less screening and inadequate follow-up. The accusations raise serious questions about whether the platform's clinical workflows were designed around patient safety or revenue throughput.
Why it matters
As telehealth becomes a primary access point for high-demand medications like GLP-1 agonists, the clinical governance structures inside these companies directly affect patient outcomes at scale.
The ReasonFirst take
This is not a story about bad actors at the margins; it is a stress test of whether asynchronous, volume-driven telehealth models can maintain the clinical standards that in-person care enforces through friction.
Sonothera is developing an ultrasound-based gene delivery platform that aims to get therapeutic genes into cells without the limitations of viral vectors. The company is targeting muscular dystrophy as an early application, with broader ambitions if the approach proves out.
Why it matters
Delivery has always been the bottleneck in gene therapy, and a non-viral, physically guided approach could change the risk and cost calculus for a wide range of diseases.
The ReasonFirst take
The delivery problem in gene therapy is real and unsolved at scale, so ultrasound-mediated transfer deserves serious attention, but the distance between a promising mechanism and a durable clinical result in muscular dystrophy is where most of these stories quietly end.
CMS has proposed increasing reimbursement by 19% for physicians who provide counseling on smoking cessation and alcohol use during patient visits. The change is designed to incentivize brief behavioral interventions that are already evidence-based but chronically underused in clinical practice.
Why it matters
Reimbursement structure is one of the most reliable levers for changing what actually happens in the exam room, and this proposal targets two of the highest-burden, most modifiable risk factors in primary care.
The ReasonFirst take
A payment nudge like this can shift documentation and billing behavior quickly, but the harder question is whether a 19% bump is enough to change the time pressure and workflow constraints that cause clinicians to skip these conversations in the first place.
MDCalc is launching a quality-rating system to evaluate the clinical calculators embedded in its platform, which is used by millions of clinicians worldwide. The ratings will assess calculators across dimensions such as evidence quality, validation, and clinical utility.
Why it matters
Clinicians rely on these tools daily for high-stakes decisions, and until now there has been no systematic, transparent way to distinguish a well-validated calculator from one built on thin evidence.
The ReasonFirst take
Rating calculators is genuinely useful, but the harder problem is whether clinicians will actually change their behavior based on those ratings, or simply keep reaching for the familiar tool regardless of its score.
The number of active medication shortages in the United States climbed to 227 in the second quarter of 2026, according to new data. The figure reflects ongoing fragility in pharmaceutical supply chains that has persisted despite years of policy attention and industry pledges to diversify manufacturing.
Why it matters
Clinicians and health system leaders are making real-time prescribing and procurement decisions under shortage conditions, and a rising shortage count means those workarounds are becoming the norm rather than the exception.
The ReasonFirst take
227 active shortages is not a supply chain blip to be managed by pharmacy teams alone; it is a systems failure that now requires health system leadership to treat shortage preparedness as a core operational competency, not a downstream logistics problem.
drug shortagespharmaceutical supply chainhealth system operationsRead
Northwest Biotherapeutics submitted its brain cancer treatment DCVax-L to the FDA over two and a half years ago and is still waiting for a regulatory decision. The prolonged review raises serious questions about FDA timelines, capacity, and what happens to patients with no other options in the meantime.
Why it matters
For clinicians and health system leaders, 931 days without a decision on a treatment for one of the most lethal cancers we manage is not a bureaucratic footnote, it is a patient care reality.
The ReasonFirst take
The real issue here is not whether DCVax-L will ultimately be approved, it is that a 931-day review clock with no clear accountability reveals a structural gap in how the FDA handles complex, high-need oncology submissions that no amount of breakthrough designation language actually fixes.
A health informatics researcher at Sacred Heart University has proposed a Family Health Record framework built on FHIR standards that would aggregate health data at the household and community level, not just the individual. The framework argues that treating the family unit as the foundational data structure could bridge the persistent gap between personal clinical care and population health management.
Why it matters
If this framework gains traction, it would require clinicians, informaticists, and health system leaders to rethink consent models, data governance, and care workflows that have been built entirely around the individual patient.
The ReasonFirst take
The concept is clinically intuitive since family context has always shaped individual health, but the hard work is not in the FHIR mapping, it is in resolving whose consent governs shared household data when family members disagree.
The FDA has approved an over-the-counter continuous glucose monitor that can be used in children as young as two years old, opening a new commercial pathway outside traditional diabetes management. Clinicians and researchers are now debating whether deploying CGM technology for childhood obesity prevention is premature, potentially harmful, or a genuine opportunity.
Why it matters
Clinician-educators and pediatric leaders will be fielding parent questions and shaping institutional guidance on a technology that is now accessible without a prescription, long before the clinical protocols to support its responsible use in young children exist.
The ReasonFirst take
Approving a device is not the same as approving a use case, and the gap between what is commercially available and what is clinically validated is exactly where pediatric patients get hurt by well-intentioned interventions.
The Trump administration is pursuing changes to the Federal Acquisition Regulation, the foundational ruleset governing all federal contracts, that would give the executive branch broader authority to direct research priorities, terminate grants, and override external scientific review processes. The move follows earlier executive actions that faced legal challenges, and this regulatory pathway is designed to be more resistant to court intervention.
Why it matters
If these changes take hold, the structural independence of federally funded research, including clinical and health services research, shifts from a norm protected by regulation to a privilege granted at executive discretion.
The ReasonFirst take
The real issue here is not any single grant termination but whether the administrative scaffolding that has kept scientific peer review insulated from political pressure can survive a deliberate, procedurally durable rewrite of the rules that hold it up.
federal research fundingscience policyexecutive authorityRead
Senior Trump administration officials including Marco Rubio, RFK Jr., and health industry figure Chris Klomp met privately with pharmaceutical industry leaders to press for onshoring of generic drug manufacturing. The push reflects growing bipartisan concern about U.S. dependence on foreign supply chains, particularly from China and India, for essential medicines.
Why it matters
Generic drugs make up roughly 90% of U.S. prescriptions, so any meaningful shift in where they are made has direct implications for drug availability, cost, and clinical reliability for every health system and prescriber in the country.
The ReasonFirst take
Onshoring generic manufacturing is a legitimate national security and supply chain priority, but the harder question no one in that room is answering publicly is who absorbs the cost increase when domestic production raises prices on the drugs that already have the thinnest margins.
Jerome Adams, former U.S. Surgeon General, published an opinion in STAT News arguing that the FDA should regulate the growing peptide market through a structured framework that centers clinician oversight. The piece comes amid a surge in direct-to-consumer peptide use, much of it operating in regulatory gray zones.
Why it matters
Clinicians are already being asked about peptides by patients who sourced them outside any formal care pathway, and the absence of a coherent regulatory structure puts prescribers in a difficult position.
The ReasonFirst take
Adams is right that clinician guidance should anchor any peptide framework. The harder, unaddressed work is the specifics: who enforces the standard, who trains the clinicians, and how compounding pharmacies fit.
A STAT News opinion piece by Daniel Morgan argues that the medical industry has prioritized developing advanced diagnostic tests without investing equally in understanding when and how to use them appropriately. Morgan contends that reducing unnecessary testing requires systemic redesign, not simply asking individual clinicians to exercise more restraint.
Why it matters
Clinician-leaders are being asked to drive value-based care while operating inside systems that make overordering the path of least resistance, and this piece names that contradiction directly.
The ReasonFirst take
We keep blaming doctors for ordering too many tests while building systems where ordering nothing requires more courage, more documentation, and more liability exposure than just running the panel.
STAT News interviewed Anthropic CEO Dario Amodei about how AI is reshaping drug discovery and development, with pharmaceutical companies increasingly integrating the technology into core workflows. The conversation surfaces real use cases beyond the usual hype cycle, pointing to molecular modeling, clinical trial design, and data synthesis as areas seeing genuine traction.
Why it matters
When the CEO of one of the most serious AI labs in the world talks specifics about biotech applications, clinicians and health system leaders need to understand what is being built upstream of them before it arrives at the bedside.
The ReasonFirst take
Pharma adopting AI for target discovery is genuinely exciting, but the leap from faster drug identification to better patient outcomes still runs straight through the messy human terrain of trial design, regulatory judgment, and clinical implementation that no model can shortcut.
A multi-institution South Korean research team, including Catholic University of Korea Seoul St. Mary's Hospital, Korea University, Asan Medical Center, and VUNO, is developing a smartphone AI application to support hospital staff in deciding whether to admit children for emergency care. The project is a direct response to paediatric specialist shortages and emergency department overcrowding.
Why it matters
Paediatric triage is one of the highest-stakes, lowest-margin-for-error decisions in emergency medicine, and any tool that gets deployed at that decision point deserves serious scrutiny from clinicians and health system leaders.
The ReasonFirst take
This is exactly the kind of AI application that sounds reassuring until you ask who carries liability when the app says discharge and the child deteriorates at home.
A Vancouver police sergeant with ALS has become the first Canadian patient to receive a Neuralink brain-computer interface implant. The procedure also marked the first use of a new experimental surgical robot designed to insert the device's electrode strands through the dura, the brain's outer protective membrane.
Why it matters
For clinicians and health system leaders, this signals that BCI technology is moving from theoretical to procedural reality faster than most care pathways, ethics frameworks, or consent processes are prepared to handle.
The ReasonFirst take
The real story here is not the implant or even the robot; it is that a police sergeant with ALS had to travel to get access to a technology that could restore communication, which tells you everything about where health systems actually stand on equitable access to emerging neurotechnology.
Anthropic has released Claude Science, an auditable AI workbench designed to support the full research lifecycle, from literature analysis to manuscript development. It runs on existing Claude models including Opus 4.8 and does not introduce new underlying AI capabilities.
Why it matters
Clinician-researchers and academic health systems evaluating AI for research workflows need to understand what they are actually buying: a structured interface, not a smarter model.
The ReasonFirst take
Claude Science is a workflow layer over existing models rather than a new scientific capability, and the real test is whether its auditability holds up when a journal asks you to defend your methods.
Sword Health has signed a partnership with Portugal's National Health Service (SNS) to make its AI-driven remote physiotherapy platform available to public system patients. Physicians in Portugal can now formally prescribe Sword's program for common musculoskeletal conditions including lower back pain, knee pain, and shoulder pain.
Why it matters
This is one of the first times an AI-enabled digital therapeutic has been integrated into a national public health system as a physician-prescribed intervention, not just a consumer add-on, which sets a precedent for how health systems elsewhere might formally adopt similar tools.
The ReasonFirst take
Getting a government health system to prescribe your product is a legitimacy milestone that no amount of venture funding can buy, but the real test is whether outcomes data from a public, resource-constrained population will hold up the way it did in employer-sponsored pilots.
musculoskeletaldigital therapeuticspublic health systemsRead
A Roche-developed KRAS inhibitor has demonstrated clinical results strong enough to be described as setting a new treatment standard for KRAS-mutant non-small cell lung cancer. This adds a credible second-generation option to a target space that only recently became druggable with the approval of sotorasib and adagrasib.
Why it matters
KRAS mutations drive roughly 25 percent of non-small cell lung cancers, and a drug that meaningfully outperforms current KRAS inhibitors reshapes how oncologists will sequence therapy for one of the most common and lethal cancers.
The ReasonFirst take
This is real signal in a space where early KRAS drugs underwhelmed, but the harder clinical question is not whether this drug works better in a trial but whether your institution will have the molecular testing infrastructure to identify the right patients before they deteriorate.
Tala Fakhouri, formerly a senior figure in FDA's AI regulatory work and now in industry, argues that biopharma companies are interpreting FDA's AI guidance more conservatively than the agency intended. She also acknowledges FDA could be doing more to reduce ambiguity and build industry confidence.
Why it matters
If companies are self-imposing restrictions that regulators never asked for, that is a compliance culture problem masquerading as a regulatory problem, and it has real costs for development timelines and AI adoption.
The ReasonFirst take
When someone who wrote the rules tells you that you are misreading them, the honest question is whether your legal and compliance teams are protecting the organization or protecting themselves.
A researcher writing in STAT News reports that the share of young people turning to AI chatbots for mental health advice increased by more than 40% in a single year. The piece calls for formal rules to protect adolescents using these tools in the absence of clinical oversight.
Why it matters
Clinicians, health systems, and policymakers are now downstream of a behavioral shift already happening at scale among the most vulnerable users.
The ReasonFirst take
Teens did not wait for us to build a framework, and the 40% jump in a single year is not a trend to monitor at the next committee meeting, it is a care delivery reality that pediatric and mental health teams need to respond to now.
The FDA has issued what is being called a historic clearance for an app that uses a large language model to help patients manage their diabetes within a treatment plan set by their physician. The clearance draws a sharp line between the LLM as a patient-facing interface and the physician as the actual decision-maker, though critics and supporters alike are questioning whether that line will hold in practice.
Why it matters
This clearance sets a regulatory and conceptual precedent that will shape how every subsequent AI-driven clinical tool is classified, marketed, and ultimately trusted by clinicians and patients.
The ReasonFirst take
The FDA has essentially blessed a legal fiction: that an LLM guiding a patient's daily diabetes decisions in real time is just an interface, not a decision-maker, and how long that fiction survives first contact with a hypoglycemic patient at 2am is the real question.
Anthropic has released Claude Science, a version of its Claude large language model optimized specifically for scientific research environments and pharmaceutical development workflows. The product targets drugmakers and lab scientists who need AI assistance tuned to the complexity and precision demands of scientific work.
Why it matters
When a frontier AI lab explicitly targets drug discovery, it signals that the pharma sector is moving from AI experimentation to infrastructure-level adoption, which has direct implications for how clinical and research leaders allocate resources and evaluate vendor partnerships.
The ReasonFirst take
A smarter autocomplete for scientists is still autocomplete, and the hard part of drug development has never been generating hypotheses; it has always been knowing which ones are worth betting a decade and a billion dollars on.
Anthropic has released Claude Science, a model variant oriented toward scientific reasoning and research tasks rather than general-purpose use. The move suggests that frontier AI labs are beginning to target the structured, evidence-dependent demands of real scientific work rather than chasing headline-grabbing demonstrations.
Why it matters
For clinicians, researchers, and health system leaders, a model built for scientific rigor rather than conversational performance changes the calculus on where AI can actually be trusted in evidence synthesis, hypothesis generation, and clinical research workflows.
The ReasonFirst take
A science-focused model is only as serious as the validation data behind it, and until Anthropic shows us benchmarks grounded in peer-reviewed scientific tasks rather than curated demos, this is a promising signal that still requires a skeptic's patience before it earns a place in any research pipeline.
A new MIT Sloan Management Review piece highlights a critical gap in corporate AI governance: while virtually every large organization claims to govern its AI systems, most leaders cannot name who holds authority to shut down a harmful model. The article argues that the real test of governance is not the existence of policies but the clarity of accountability when something goes wrong.
Why it matters
In healthcare and other high-stakes sectors, the gap between having a governance framework and having a named, empowered person who can act in a crisis is exactly where patient safety risk lives.
The ReasonFirst take
Governance is not the document; it is a named decision-maker with the authority and a tested process to shut a model down. Without that, the policy is oversight in name only.
Anthropic launched Claude Science, a research platform for life sciences, joining Google DeepMind and OpenAI in building AI tools aimed at drug discovery and scientific work.
Why it matters
When frontier labs target science directly, AI stops being a chat assistant and becomes a research collaborator that has to be supervised.
The ReasonFirst take
The tool to watch is not the chatbot; it is the lab assistant. When frontier labs target science directly, the question for programs is who learns to supervise it.
A June 2026 Nature Medicine benchmark found frontier general models outperformed FDA-cleared clinical AI tools on real physician queries, raising a question regulators have not answered: does a cleared device actually beat the free model a clinician already has?
Why it matters
Clearance signals safety, not superiority. If a cleared tool underperforms the model a clinician already has open, clearance is measuring the wrong thing.
The ReasonFirst take
Clearance tells you a tool cleared a bar. It does not tell you it beats what you already have in another tab. Ask for the head-to-head, not the badge.
At the 2026 MIT Sloan CIO Symposium, technology and business leaders reflected on their first serious encounters with agentic AI in production environments. The consistent finding was a gap between what vendors and early demos promised and what organizations actually experienced when agents touched live workflows.
Why it matters
If agentic AI is already inside operational workflows at leading organizations, clinical and health system leaders cannot afford to treat this as a future-state conversation.
The ReasonFirst take
The more important question buried in this piece is not whether the agents are ready, but whether organizations have done the harder work of clarifying which decisions actually require human judgment before they hand anything to an agent.
Generate's GB-0895, an anti-TSLP antibody for severe asthma, became the first AI-derived antibody to reach Phase III, one of an estimated 15 to 20 AI programs entering pivotal trials this year. Analysts frame 2026 as validation and disappointment in roughly equal measure.
Why it matters
AI has compressed discovery, but the clinic is where the claim is finally tested, and that timeline has not changed.
The ReasonFirst take
Faster to the clinic is not faster to proof. The molecule was designed by a model; the answer still comes from a trial, at the speed a trial takes.
McKinsey finds just 21% of companies have redesigned workflows end to end and only 1% describe their AI as mature, while leaders estimate employee AI use at roughly a third of the real rate.
Why it matters
The bottleneck is not access to tools. It is the willingness to redesign the work around them, which is a leadership decision.
The ReasonFirst take
The gap is not adoption, it is redesign. Tools land on desks; the work never changes shape. Maturity is a workflow question, not a software one.
A 2026 npj Digital Medicine systematic review and meta-analysis finds human-LLM collaboration can enhance clinical reasoning, documentation, and interpretation, with the benefit depending on how the collaboration is structured, not the model alone.
Why it matters
The value is not in the model. It is in the design of the human-machine pairing, which is exactly what can be taught and standardized.
The ReasonFirst take
The finding is not 'AI helps.' It is that the pairing has to be designed. Trust calibration is the variable, and it is teachable.
SHRM finds 41% of workers now use AI at work, US postings requiring AI skills grew 144% year over year, and AI-skilled workers command wage premiums up to 56%. 84% of CHROs expect AI-upskilling demand to rise.
Why it matters
The labor market is repricing around the ability to work with AI, and the premium goes to judgment about the machine, not mere use of it.
The ReasonFirst take
The premium is not for using AI. It is for managing it. The teachable skill is judgment about the machine, and the market is now pricing it in.
Q1 2026 funding reached $4B across 110 deals, with 59% concentrated in 12 mega-deals. Rock Health retired its separate 'AI deal' tracking because AI has become table stakes in how digital health companies are built.
Why it matters
When AI stops being a distinct funding category, the market is telling you it is now a baseline expectation, not a differentiator.
The ReasonFirst take
When 'AI company' stops being a category, the question stops being whether a tool uses AI and becomes whether it removes real work.
Programs are shifting from multiple-choice toward assessing a student's reasoning process in simulated settings, and adding explicit AI-literacy coursework, as generative AI makes the recalled answer trivial to produce.
Why it matters
If a student can generate the answer instantly, the assessment has to test the thinking behind it, not the answer itself.
The ReasonFirst take
If the model can produce the answer, the answer was never the skill. Assess the defense of the reasoning, not the output.
A Lancet Public Health agenda for policymakers argues AI can improve population health and equity only if built on robust public health infrastructure, and warns it may otherwise widen the very disparities it promises to close.
Why it matters
AI amplifies whatever system it sits on. On a fragmented public health system, it amplifies the fragmentation.
The ReasonFirst take
AI does not fix a system it is layered onto. Bolt it onto a fragmented public health system and you automate the fragmentation.
The FDA will exercise enforcement discretion for clinical decision support tools that give a single reviewable recommendation, and widened its wellness carve-out for consumer wearables. Enforcement of the new documentation regime begins June 30, 2026.
Why it matters
When the regulator steps back, the burden of judging whether a tool is safe and appropriate shifts onto the clinician and the institution.
The ReasonFirst take
Less oversight moves the burden of proof from the regulator to you. The clinician is now the validation study.
A small but growing set of digitally mature health systems are adding equity-monitoring requirements to AI procurement, an early sign that fairness is becoming a purchasing criterion rather than a value statement.
Why it matters
A value only changes behavior once it is written into the contract. Procurement is where transformation is actually enforced.
The ReasonFirst take
Values become real when they enter the contract. Watch procurement, not mission statements; that is where transformation is enforced.